MODULE 6

 Planning in AI
Introduction:
 Planning in artificial intelligence involves decision-making
processes executed by robots or computer programs to achieve
predefined goals.
 It is a critical aspect of AI projects, essential for tasks ranging
from route optimization to complex problem-solving.

1. Forward State Space Planning (FSSP):

 FSSP operates similarly to forward state-space search.
 It begins with an initial state and progresses through a
sequence of actions to reach the target state.
 Each action is applied to the current state, leading to a new
state, until the goal state is achieved.
 Advantage: Sound algorithm ensuring correctness.
 Disadvantage: Large branching factor, which can lead to
computational complexity.

2. Backward State Space Planning (BSSP):

 BSSP functions related to backward state-space search.
 It starts from the goal state and traces back through a
series of actions to identify sub-goals or preconditions
necessary for achieving the final goal.
 This process, known as regression, involves moving from
the target state to the sub-goals.
 Advantage: Smaller branching factor compared to FSSP,
reducing computational complexity.
 Disadvantage: Not always sound, as inconsistency may
arise.

Role of Planning in AI:

 Planning is essential for decision-making and action selection in
AI systems.
 It involves determining the sequence of tasks to be performed
by the system to accomplish a specific goal.
 Effective planning ensures efficient utilization of resources and
optimal task execute
 Types of planning
1. Classical Planning:
 Classical planning involves creating a sequence of actions
to achieve a predefined goal within a static and
predictable environment.
 Assumes a deterministic world where outcomes of actions
are known in advance and do not change over time.
 Example: Planning a route from point A to point B using a
map where the road network is fixed.
2. Hierarchical Planning:
 Hierarchical planning breaks down complex problems
into smaller, more manageable sub-problems, organized
in a hierarchical structure.
 Involves creating a hierarchy of plans, with higher-level
plans overseeing the execution of lower-level plans.
 Example: Planning a trip involves higher-level plans such
as booking flights and accommodations, which supervise
lower-level plans like packing luggage and arranging
transportation.
3. Temporal Planning:
 Temporal planning considers time constraints and
interdependencies between actions when creating plans
for the future.
 Ensures that plans are executable within specific time
limits by accounting for the duration of tasks and their
temporal relationships.
 Example: Scheduling tasks in a project management tool,
where dependencies and deadlines dictate the order and
timing of activities.

 Components of the planning system:

1. Rule Selection: Choose the most appropriate rule based on
available information and heuristics to guide the planning
process effectively.
2. Rule Application: Apply the selected rule to update the
problem condition, moving closer to the desired goal state.
3. Solution Detection: Identify when a solution has been found by
reaching the target state or achieving specified criteria.
 4. Dead-End Detection: Detect dead ends or infeasible paths to
discard them and focus efforts on more promising directions.
5. Optimality Detection: Determine when a near-perfect solution
is reached, balancing between computational resources and
solution quality.

 Block-world planning problem:

 The block-world problem, also known as the Sussmann

anomaly.
 It involves manipulating blocks labeled 'A', 'B', and 'C' on
a flat surface to achieve a target configuration, with the
constraint that only one block can be moved at a time.

Problem Description:
 Start Position: Blocks arranged in a specific
configuration on the flat surface.
 Target Position: Desired arrangement of blocks to be
achieved through a series of movements.
  Only one block can be moved at a time, and each
movement must be carefully planned to reach the target
configuration.

Challenge:
 Non-interleaved planners of the early 1970s struggled
to solve the block-world problem efficiently due to its
complexity.
 When faced with two sub-goals (G1 and G2), these
planners often failed to produce a combined plan for
both sub-goals, leading to an odd behavior.

 Goal stack planning:

Blocks World Problem:

The Blocks World Problem involves manipulating blocks
on a table using a robot arm to achieve a desired
configuration. The scenario typically includes:
 Table: A flat surface where blocks can be placed.

 Blocks: Objects that can be moved and stacked on the

table.
 Robot Arm: A mechanism capable of picking up and

putting down blocks one at a time.

 Constraints: Only one block can be moved at a time,
and no block should be stacked on top of another
during movement.

Goal Stack Planning:

Goal Stack Planning is an AI problem-solving method that
works backward from the goal state to the initial state.
Here's how it works:
1. Initialization: Start with the goal state and work
backward toward the initial state.
2. Break down the goal into subgoals and recursively solve
them.
3. Select actions to achieve subgoals based on
preconditions and effects.
4. Execution: Execute selected actions to transform the
world state closer to the goal state.
5. Backtracking: If a deadlock or inconsistency is
encountered, backtrack and explore alternative paths.
6. Termination: Repeat steps until the initial state is
reached, and the goal is achieved.

Components of Goal Stack Planning:

 Initial State: The starting configuration of blocks.

 Goal State: The desired configuration to be achieved.

 World State: Represents the current state of the

environment during planning.

 Predicates: Statements conveying information about

the configuration in the Blocks World, such as block

positions and robot arm status.
 Operations: Actions performed by the robot arm,

including picking up, putting down, stacking, and

unstacking blocks.

Representing Configurations with Predicates:

 - ON(A, B): Block A is stacked on top of block B.
- ONTABLE(A): Block A is placed directly on the table.
- CLEAR(A): There is no other block stacked on top of
block A.
- HOLDING(A): The robot arm is currently holding
block A.
- ARMEMPTY: The robot arm is not holding any
block.
Using these predicates, we can describe both the initial
and goal states:
 Initial State:
ON(B, A) ∧ ONTABLE(A) ∧ ONTABLE(C) ∧ ONTABLE(D)
∧ CLEAR(B) ∧ CLEAR(C) ∧ CLEAR(D) ∧ ARMEMPTY

 Goal State:
ON(C, A) ∧ ON(B, D) ∧ ONTABLE(A) ∧ ONTABLE(D) ∧
CLEAR(B) ∧ CLEAR(C) ∧ ARMEMPTY
Operations of the Robot Arm:
The robot arm can perform four operations, each with
specific preconditions:
- STACK(X, Y): Stacking block X on top of block Y.
o Preconditions: CLEAR(X), HOLDING(X), CLEAR(Y)
- UNSTACK(X, Y): Picking up block X from the top of
block Y.
o Preconditions: ON(X, Y), CLEAR(X), ARMEMPTY
- PICKUP(X): Picking up block X from the table.
o Preconditions: ONTABLE(X), CLEAR(X),
ARMEMPTY
- PUTDOWN(X): Placing block X on the table.
o Preconditions: HOLDING(X)

These operations enable the robot arm to manipulate

blocks and rearrange them to achieve the desired goal
state. Each operation requires certain conditions to be
met before it can be executed successfully, ensuring the
safety and effectiveness of block movements.

Solution:
- The effect of these operations is represented using two
lists ADD and DELETE.
- DELETE List contains the predicates which will cease to
be true once the operation is performed.
 - ADD List on the other hand contains the predicates
which will become true once the operation is
performed.

 Non-linear planning:
 Non-Linear Planning in AI is a problem-solving approach used
to address complex situations with unpredictable solutions.
 It involves finding a sequence of actions to transform the
system's current state to the desired goal state, considering non-
linear constraints like resource limitations and temporal
ordering.
 - Algorithm:
1. Select Goal:
 Choose a goal, denoted as 'g', from the set of goals.
2. Operator Selection:
 If the current state does not match goal 'g', select an
operator 'o' whose add-list matches goal 'g'.
3. Push Operator:
 Push operator 'o' onto the OpenStack and add the
preconditions of 'o' to the set of goals.
4. Continue Process:
 Continue this process while all preconditions of the
operator on top of the OpenStack are satisfied in the
current state.
5. Apply Operator:
 Pop the operator 'o' from the top of the OpenStack, apply it
to the current state, and add it to the plan.

Advantages:
 Flexibility: Handles complex problem spaces with
interdependent sub-goals and non-linear dependencies.
 Optimality: Utilizes heuristic search techniques to find optimal
solutions.
 Resource Utilization: Considers resource limitations and
temporal ordering constraints.
 Reusability: Can reuse previously generated plans for similar
problems, improving efficiency.
 Integration: Can be integrated with other AI techniques like
machine learning for robust problem-solving.

Disadvantages:
 Complexity: Dealing with complex and interconnected sub-
goals can increase computational overhead.
 Scalability: May face challenges with large problem spaces,
where the search space becomes too large to explore
efficiently.
 Uncertainty: Assumes a deterministic world, which may not
always hold true in real-world scenarios.
 Domain Expertise: Relies on domain expertise to generate
effective plans, which may require significant knowledge.